Children (n = 122) and adults (n = 200) with dyslexia completed rapid automatic naming (RAN) letters, rapid automatic switching (RAS) letters and numbers, executive function (inhibition, verbal fluency), and phonological working memory tasks. Typically developing 3rd (n = 117) and 5th (n = 103) graders completed the RAS task. Instead of analyzing RAN/RAS results the usual way (total time), growth mixture modeling assessed trajectories of successive times for naming 10 symbols in each of five rows. For all three samples and both RAN and RAS, two latent classes were identified. The "faster" class performed slowly on the first row and increased time by "small increments" on subsequent rows. The "slower" latent class performed "more slowly" on the first row, and children, but not adults, increased time by "larger increments" on subsequent rows. For children, both the initial row (automaticity index) and slope (sustained controlled processing index) of the trajectory differentiated the classes. For adults, only the initial row separated the classes. The longest time was on row 3 for RAN and row 4 for RAS. For the typically developing 5th graders, close in age to the children with dyslexia, the trajectories were flatter than for children with dyslexia and only the slower class (4%) showed the peak on row 4. For children with dyslexia, inhibition predicted RAN slope within the slower latent class and phonological working memory predicted RAS slope for both latent classes. For adults with dyslexia, inhibition and phonological working memory differentiated both latent classes on RAN intercept and RAS slope. Taken together, RAN, which may assess the phonological loop of working memory, and RAS, which may assess the central executive in working memory, may explain the timing deficit in dyslexia in sustaining coordinated orthographic-phonological processing over time.